I'd like to engrave music with quarter-tones in Lilypond, and generate MIDI that will sound correct when played. Lilypond is supposed to support that but I've run into some issues with actually doing it.
First: quarter tones are implemented by emitting the nearest 12-ET note and a quarter-tone pitch bend if applicable, so that for instance cih' (quarter tone sharper than Middle C) is MIDI note 60 (c') with a +50 cent pitch bend. And MIDI pitch bends affect the entire channel; so if there are two notes that require different bend amounts, then they have to be in different MIDI channels. I can basically resolve this issue using code posted to the list by Jean Abou Samra, which uses a "performer" to allow querying out single lines from simultaneous notes into separate channels: https://lists.gnu.org/archive/html/lilypond-user/2023-05/msg00324.html A disadvantage of this approach is that it requires a separate channel for each simultaneous note, when it would seem that two channels (one for no bend, one for a quarter-tone bend) should be enough. However, other issues remain. The second one is that Lilypond sometimes emits a message cancelling the pitch bend at the same moment as the Note Off message - which is not actually the end of the note, only the start of the "release" of the note. So the MIDI renderer obeys the pitch bend message at that moment, and the note ends up with its release bent to the wrong pitch. I've not been able to determine exactly when Lilypond does or doesn't do this. It happens on some notes and not others. I also am not sure whether the note-splitting code used for the first issue may be having some effect on it. And in the case of legato playing, even if Lilypond never "cancelled" the pitch bend at the Note Off message, it might still need to apply a different pitch bend for the next note before the release of the previous note has completed, so just eliminating "cancel pitch bend" at Note Off would not be enough to resolve it anyway. Ideally, one would split notes not according to simultaneity but according to amount of pitch bend. There could be one channel permanently set with no pitch bend, and a second channel (only two channels total) permanently set to a quarter tone pitch bend, and notes somehow filtered or routed to those channels according to their pitch. This might not work for every "microtonal" scale, because there could be many distinct pitch bend values in play, but my particular case of 24-ET is easy because there are really only two pitch bend values ever needed. That raises the third issue: Lilypond doesn't actually use just two pitch bend values for quarter tones. For a note that is not a straight 12-ET MIDI note, Lilypond either uses the next lower 12-ET note, plus 50 cents; or the next higher 12-ET note minus 50 cents. I have not been able to determine how it decides this. At first I thought it was note spelling (cih' becoming note 60 plus 50 cents, deseh' becoming note 61 minus 50 cents) but in further testing, that doesn't seem to be the case. I haven't been able to rule out that it may be some consistent function of the pitch (that is, I haven't found examples of the same pitch resulting in two different pitch bends, from two different starting note numbers) but it's definitely the case that over the whole scale the set of pitch bends used is not minimal. It uses all three of 0, -50, and +50 even though it would be possible to express all 24-ET pitches using only two of those. I'd like to be able to enter music in Lilypond so that it engraves right, and then filter it (either in Lilypond or some postprocessing step) so that notes end up in two MIDI channels, one of which has permanently no pitch bend and the other of which has permanently +50 cent pitch bend. There should be no change in a channel's pitch bend during the music (thus, not during decay tails of notes), and every note should play at its correct pitch without being bent to a different pitch during release. It would be preferable that this end up involving only two output channels: not three, not one per simultaneous note, and not extra channels even beyond that. Any thoughts on how to achieve this? It seems like I could get most of the way by tagging and filtering on note names and quarter-tone accidentals, to get all notes that are 12-ET notes into one channel for "no bends" and all notes that aren't 12-ET notes into a another, permanently-bent channel. Then I can easily enough filter out any spurious "cancel bend" messages because I know each channel will always have a fixed bend amount. But even if I'm willing to use three output channels instead of two, I need to know what actually causes Lilypond to choose +50 or -50 bends, to decide into which channel a given note should go so that the bend amount for each channel will be the same for all notes in that channel. -- Matthew Skala msk...@ansuz.sooke.bc.ca People before tribes. https://ansuz.sooke.bc.ca/
